Datasets:
Upload fiftyone.yml with huggingface_hub
Browse files- fiftyone.yml +16 -0
fiftyone.yml
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
description: VisualOverload (CVPR 2026) is a visual question answering (VQA) benchmark
|
| 2 |
+
of 2,720 question-answer pairs over 150 high-resolution public-domain paintings
|
| 3 |
+
densely populated with figures, actions, and unfolding subplots. It probes whether
|
| 4 |
+
VLMs can perform simple, knowledge-free vision tasks in overloaded scenes. Ground-truth
|
| 5 |
+
answers are held privately; models are scored via the official evaluation server
|
| 6 |
+
using each question's question_id. This is a FiftyOne-format version of paulgavrikov/visualoverload.
|
| 7 |
+
format: FiftyOneDataset
|
| 8 |
+
license: cc-by-sa-4.0
|
| 9 |
+
name: VisualOverload
|
| 10 |
+
tags:
|
| 11 |
+
- fiftyone
|
| 12 |
+
- fiftyone
|
| 13 |
+
- image
|
| 14 |
+
- image
|
| 15 |
+
- visual-question-answering
|
| 16 |
+
- vqa
|